An Early Performance Evaluation of the NEXRAD Dual-Polarization Radar Rainfall Estimates for Urban Flood Applications
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
1478 WEATHER AND FORECASTING VOLUME 28
An Early Performance Evaluation of the NEXRAD Dual-Polarization Radar Rainfall
Estimates for Urban Flood Applications
LUCIANA K. CUNHA
Department of Civil and Environmental Engineering, Princeton University, Princeton, New Jersey, and Willis Research
Network, London, United Kingdom
JAMES A. SMITH AND MARY LYNN BAECK
Department of Civil and Environmental Engineering, Princeton University, Princeton, New Jersey
WITOLD F. KRAJEWSKI
IIHR—Hydroscience and Engineering, The University of Iowa, Iowa City, Iowa
(Manuscript received 9 April 2013, in final form 24 July 2013)
ABSTRACT
Dual-polarization radars are expected to provide better rainfall estimates than single-polarization radars
because of their ability to characterize hydrometeor type. The goal of this study is to evaluate single- and dual-
polarization radar rainfall fields based on two overlapping radars (Kansas City, Missouri, and Topeka,
Kansas) and a dense rain gauge network in Kansas City. The study area is located at different distances from
the two radars (23–72 km for Kansas City and 104–157 km for Topeka), allowing for the investigation of radar
range effects. The temporal and spatial scales of radar rainfall uncertainty based on three significant rainfall
events are also examined. It is concluded that the improvements in rainfall estimation achieved by polari-
metric radars are not consistent for all events or radars. The nature of the improvement depends funda-
mentally on range-dependent sampling of the vertical structure of the storms and hydrometeor types. While
polarimetric algorithms reduce range effects, they are not able to completely resolve issues associated with
range-dependent sampling. Radar rainfall error is demonstrated to decrease as temporal and spatial scales
increase. However, errors in the estimation of total storm accumulations based on polarimetric radars remain
significant (up to 25%) for scales of approximately 650 km2.
1. Introduction 2008; Villarini and Krajewski 2009). These uncertainties
limit the use of SPR QPEs as input to flood simulation
Weather radars have been successfully used for many
models, especially for urban basins with rapid responses
years to monitor rainfall and extreme weather events. In
to short-term rainfall rates (e.g., Carpenter and Geor-
the United States, the first Weather Surveillance Radar-
gakakos 2004; Javier et al. 2007; Smith et al. 2007;
1988 Doppler (WSR-88D) was installed in 1992, and
Schr€oter et al. 2011; Cunha et al. 2012).
since then 159 single-polarization S-band radars (SPRs)
One of the major sources of uncertainty in SPR rainfall
have been deployed. These instruments efficiently de-
estimates is the lack of a unique relationship between
tect the structure and evolution of storms and provide
reflectivity, measured by the radar, and rainfall rate. For
quantitative precipitation estimates (QPEs). There are,
SPR QPEs, measurements of reflectivity are converted
however, large uncertainties in these QPEs (Wilson and
into rainfall rate (mm h21) by a power-law relationship
Brandes 1979; Fabry et al. 1992; Baeck and Smith 1998;
(Z 5 aRb) for which parameters are empirically esti-
Smith et al. 1996; Germann et al. 2006; Szturc et al.
mated. Since the parameters of this relationship depend
on the raindrop size distribution, which varies between
and within storms, there is no unique Z–R relationship
Corresponding author address: Luciana Cunha, Department
of Civil and Environmental Engineering, Princeton University,
that can satisfy all meteorological phenomena. Other
E-208, Engineering Quad, Princeton, NJ 08544-0001. factors that lead to inaccuracies in radar-derived rainfall
E-mail: lcunha@princeton.edu estimates include bright bands (maximums in the radar
DOI: 10.1175/WAF-D-13-00046.1
Ó 2013 American Meteorological Society
DECEMBER 2013 CUNHA ET AL. 1479
reflectivity caused by melting snow), incomplete beam This works centers on the following questions: Are
filling, calibration errors, attenuation, anomalous prop- rainfall estimates based on DPRs (consistently) better
agation (AP), ground clutter, and the sampling strategy than rainfall estimates based on SPRs? If not, in which
used by the radar (e.g., Austin 1987; Fo et al. 1998; situations are they better or worse? What is the effect of
Villarini and Krajewski 2009). range on radar rainfall estimation based on DPRs and
Upgrade of the Next Generation Weather Radar SPRs? Do DPRs provide better estimates of hydrolog-
(NEXRAD) network to dual-polarization capabilities ically relevant rainfall properties that lead to improved
began in 2011. DPRs have the potential to mitigate some prediction of flash floods?
of the uncertainties common to SPRs, since they are able In the following section, we present the study area and
to discriminate between hydrometeors and other aerial experimental data. In section 3, we present the analysis
targets, and to better characterize the hydrometeor type, methods applied in this study. In section 4, we examine
as well as the raindrop size distribution illuminated by the main results; and in section 5, we discuss our con-
the radar beam. SPR provides only measurements of clusions and recommendations for future research.
horizontal reflectivity, and DPR provides both vertical
and horizontal polarization measurements, which are
2. Study area, data, and methods
used to estimate differential reflectivity, differential phase
shift, and the copolar correlation coefficient [see Straka In this study we compare rainfall fields obtained by
et al. (2000) for definitions]. These additional variables SPRs and DPRs with ground reference data obtained
are used to improve the QPEs. Previous studies have from a dense rain gauge network in the Kansas City
demonstrated the ability of DPRs to classify hydrome- metropolitan area. We analyze radar rainfall estimates
teor types (HTs), to identify the melting layer position for three rainfall events that occurred on 19–20 March
(Straka et al. 2000; Heinselman and Ryzhkov 2006; 2012, 6–7 May 2012, and 31 August–1 September 2012.
Giangrande et al. 2008; Park et al. 2009), and to improve Each of the storms occurred after the implementation of
rainfall estimation (Chandrasekar et al. 1990; Ryzhkov the DPR upgrades to the KEAX (10 February 2012) and
et al. 2005b; Vulpiani and Giangrande 2009). KTWX (30 January 2012) WSR-88Ds. The three storms
To make the best use of these instruments, we need exhibited very different characteristics in terms of the
to fully understand the benefits they provide for QPEs spatial variability of rainfall over the study region, the
under different meteorological conditions. This is es- magnitudes of the rainfall rate, and the distribution of
pecially true for applications to urban flooding for which hydrometeor types. We will describe these differences in
high temporal and spatial resolutions, as well as accurate the results session.
rainfall estimates, are required (Krajewski and Smith
a. Rain gauge network
2002; Smith et al. 2002; Einfalt et al. 2004; Wright et al.
2012). In this study we evaluate SPR and operational The city of Overland Park and Johnson County,
DPR QPE products collected by the Kansas City, Missouri Kansas, maintain a dense rain gauge network with the
(KEAX), and Topeka, Kansas (KTWX), radars over the goal of mitigating urban floods. The network collects
Kansas City metropolitan region. We use Hydro- rainfall data in near–real time with high temporal reso-
NEXRAD to estimate SPR QPEs, while DPR QPEs lution (less than 5 min) and makes the data freely avail-
were obtained from the instantaneous precipitation rate able (www.stormwatch.com). The network includes 136
(IPR) product of the operational NEXRAD algorithms tipping-bucket rain gauges that were operational for the
and provided by the National Oceanic and Atmospheric year 2012; most of the gauges are High Sierra model
Administration (NOAA). Both radars overlap an area 2400 gauges (M. Ross 2012, personal communication).
with a very dense network of rain gauges. Our goal is to The bucket volume is 1 mm and the collected data are
better understand QPE improvements achieved by DPRs, transmitted to the base server in near–real time. While
and how they can improve flood prediction across mul- the bucket size is too coarse for detailed studies of rainfall
tiple scales. rate variability (see Habib et al. 2001), it is deemed ad-
We present an error analysis of QPEs by comparing equate for high-intensity and accumulation events that
the different radar rainfall products with rain gauge produce flash floods.
data. We evaluate possible causes of errors in SPR and We collected raw data in breakpoint format (times of
DPR QPEs and assess how QPE uncertainties vary 1-mm tips) and estimated 15-min rain rates using the tip
across a range of temporal and spatial scales. Our focus interpolation method. Ciach (2003) demonstrated the
is on urban flooding, and our evaluation is for basins with advantage of the tip interpolation method compared
drainage areas varying from approximately 1 to 650 km2 to the traditional tip-counting method, especially for
in the Kansas City area. shorter time scales. Even though we use gauge-based
1480 WEATHER AND FORECASTING VOLUME 28
estimates as ground reference, we recognize that tipping- for single-polarization (conventional NEXRAD sys-
bucket rain gauge data are subject to systematic and ran- tem) data and KEAX-DPR and KTWX-DPR for dual-
dom instrumental errors (Zawadzki 1975; Austin 1987; polarimetric data.
Habib et al. 2001; Lanza and Stagi 2008). For the generation of the 15-min SPR NEXRAD
The accuracy of the gauge versus radar comparisons is rainfall fields we collected level II data from the Na-
limited by the space–time sampling differences of both tional Climatic Data Center (NCDC) for the two radars.
instruments. While tipping buckets sample areas on the We used the Hydro-NEXRAD radar-rainfall estimation
order of 1021 m2, radar samples areas of approximately algorithms to convert reflectivity into 15-min rainfall
1 3 106 m2. The severity of the effect depends on the accumulation fields at 1-km horizontal resolution. Hydro-
spatial variability of rainfall at a given temporal scale. NEXRAD capabilities and the procedures used to
Due to the small sample of storms available for this generate rainfall products are described in Krajewski
study, we did not pursue the error variance and co- et al. (2011a) and Krajewski et al. (2007). The prepro-
variance separation method (Ciach and Krajewski 1999; cessing rain-rate algorithm converts reflectivity from
Mandapaka et al. 2009) that ‘‘isolate’’ errors due to the lowest elevation angle scan to rainfall rate using
sampling scale differences. Nevertheless, the radar versus the standard Z–R relationship for convective events
gauge comparison does provide diagnostic information (Fulton et al. 1998). The user can specify the power-law
pertinent to our goal of comparing SPR and DPR pre- empirical relationship between the reflectivity and
cipitation estimates. The scale problem is especially rel- rainfall intensity. For this study we applied the ‘‘quick
evant in the comparison of 15-min gauge and radar rain look’’ algorithm (refer to Krajewski et al. 2011a), which
rates, but it is minimized as we aggregate the observations provides standard parameters (Z 5 300R1:4 ) to gener-
in space or time (Ciach and Krajewski 1999; Seo and ate 15-min and 1 km 3 1 km accumulated rainfall maps.
Krajewski 2010; Seo and Krajewski 2011). We examine We applied a quality control algorithm that performs
both storm total analyses (aggregation in time) and basin- anomalous propagation correction (Steiner and Smith
scale error analyses (aggregation in space). 2002), but omits range correction and advection cor-
rection. Hydro-NEXRAD provides rainfall fields with
b. Radar rainfall data
a nominal grid resolution of 1 km 3 1 km in the Super
In this study we developed SPR and DPR rainfall Hydrologic Rainfall Analysis Project (SHRAP) format
fields over an area of approximately 3600 km2 (60 km by based on the HRAP algorithm of Reed and Maidment
60 km centered) around Kansas City, using observations (1999). We used standard procedures to reproject the
from both the KEAX and KTBW radars. The distance dataset to geographic coordinates.
from the KEAX radar to the rain gauges in the Kansas In this study, we utilized the operational instan-
City network ranges from 23 to 72 km, while the distance taneous precipitation rate (IPR) product to derive the
from the KTWX radar ranges from 104 to 157 km. The DPR rainfall fields for both KEAX and KTBW. The
difference in distance from both radars to the study area DPR rainfall algorithms (Istok et al. 2009) were devel-
allows us to investigate the effects of range on QPE oped in part from the results obtained by the Joint
uncertainty. Range effects on radar rainfall estimation Polarization Experiment (JPOLE) described by Ryzhkov
were previously demonstrated by Smith et al. (1996), Fo et al. (2005b), Heinselman and Ryzhkov (2006), and Park
et al. (1998), Fulton et al. (1998), and Krajewski et al. et al. (2009).
(2011b). One of the advantages of DPR is the classification of
Figure 1a shows the locations of the two radars and the hydrometeor type in the sample volume. In addition
their distances to the study area, Fig. 1b presents a map to the IPR rainfall fields, we also use the operational
with the rain gauge locations, and Fig. 1c illustrates a hybrid hydrometeor classification (HHC) fields. For the
vertical profile where we indicate the radar beam ge- initial operational capabilities the NEXRAD agencies
ometry and height over the study area for both radars for defined the following hydrometeor classes: 1) biological,
the lowest elevation angle. Range effects are caused 2) ground clutter and anomalous propagation, 3) ice
mainly by an increase in the radar beam height and radar crystal, 4) dry snow, 5) wet snow, 6) light–moderate rain,
echo vertical structure heterogeneity with range. De- 7) heavy rain, 8) big drops, 9) graupel, 10) hail (mixed
pending on the range, the radar beam might partly in- with rain), and 11) unknown. The HTs are defined based
tercept the melting layer or even completely overshoot on a fuzzy logic method that takes into consideration
precipitation. measurement uncertainties: lower weights are given to
For both radars we will compare rainfall fields that are uncertain measurements in the classification scheme.
based on the SPR-only and DPR NEXRAD measure- The melting layer (ML) position is usually identified based
ments. Herein, we refer to KEAX-SPR and KTWX-SPR on the layer with maximum reflectivity and differential
DECEMBER 2013 CUNHA ET AL. 1481
FIG. 1. (a) The 200-km umbrellas of the KEAX and KTWX radars and the basin location. (b) Rain gauge locations
(blue dots) with gray shading denoting different degrees of the Kansas City impervious area (2006). (c) A schematic
vertical profile where the KEAX and KTWX sampling areas (beam geometry and height) above the basin are indicated.
reflectivity, and minimum correlation coefficient (Brandes Table 1 presents a summary of the hydrometeor
and Ikeda 2004; Giangrande et al. 2008). Based on these classes and estimation equations used. These equations
parameters a snow score is estimated and used to define are defined by Giangrande and Ryzhkov (2008) based
the melting-layer top and bottom. When the melting on empirical data for Oklahoma. They note that the
layer cannot be identified by this procedure, the top of parameters for these equations are optimized for the
the melting layer is defined at the 08C isotherm and the Oklahoma data and that the validity of the equations
bottom is located 500 m below the top. The 08C isotherm should be tested using data collected under different
is obtained from the Rapid Update Cycle model [see climatological conditions.
Benjamin (1989) and Schuur et al. (2011) for details]. Differential reflectivity Zdr and reflectivity Z are
Once the height of the melting layer is defined, the HHC used in the case of light, moderate, and heavy rain, and
is estimated based on the best–lowest available scan big drops. In the case of hail below the melting layer,
for each location. The HHC and the melting layer po- only DZ is used. For the remaining cases (wet snow,
sition determine the rainfall-rate estimation algorithm graupel, hail above the melting layer, dry snow, and ice
(Giangrande and Ryzhkov 2008). Note that the different crystals), the standard Z–R relationship is used and
procedures used to define the melting layer can be a fixed correction is performed depending on the class
a source of uncertainty on the definition of HHC and (Ryzhkov et al. 2005b). For example, for wet snow the
consequently on QPE estimation. value calculated by the standard Z–R relationship is
1482 WEATHER AND FORECASTING VOLUME 28
TABLE 1. Hydrometeor types and radar rainfall estimation equations.
Symbol Hydrometeor type Position in relation to the ML* Rain equation
GC Ground clutter–anomalous propagation CB, PB, PA, MW R50
BI Biological CB, PB, MW, PA R50
RA Light–moderate rain CB or PB R(Z, Zdr)
HR Heavy rain CB or PB R(Z, Zdr)
BD Big drops CB, PB, MW, PA R(Z, Zdr)
HA Hail mixed with rain CB, PB, MW, PA R(Zdr)
CA 0.8R(Z)
DS Dry snow CB, PB, MW, PA R(Z)
CA 2.8R(Z)
WS Wet snow PB, MW, PA 0.6R(Z)
GR Graupel PB, MW, PA, CA 0.8R(Z)
IC Ice crystal CA 2.8R(Z)
UK Unknown All R(Z)
* CB 5 completely below, PB 5 partly below, MW 5 mostly within, PA 5 partly above, and CA 5 completely above.
reduced by 40%, and for graupel and hail by 20%. For available for polarimetric radars. For all the analyses
dry snow and ice crystals, the Z–R values area increased performed in this paper, we calculate the normalized bias
by 280%. (NB); the standard error (SE), also called root-mean-
Some of the HHC classes are identified only if the square error; and the Pearson correlation coefficient
radar beam partly (or completely) overshoots the rain- (CORR), which are defined as follows:
fall and intercepts parts of the melting layer. For ex-
normalized bias,
ample, ice crystals are observed if the radar beam
completely overshoots the melting layer. Graupel, dry
åD(s,t) 2 Dref(s,t)
snow, and wet snow are typically all present when part of NB(s) 5 ;
the radar beam is actually sampling the melting layer. åDref(s,t)
We use these classes to identify the regions in the study
area for which either the KEAX or KTWX radar standard error,
overshoots the rainfall. Overshooting is more likely to sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
happen for KTWX since it is located farther from the å[D(s,t) 2 Dref(s,t) ]2
study area. SE(s) 5 ; and
N
The IPR fields are in a polar coordinate system and
we used the NOAA toolkit to transform the data to Pearson correlation coefficient,
geographical coordinates, with spatial resolution of ap-
proximately 1 km 3 1 km. Both IPR and HHC fields are cov[D(s,t) , Dref (s,t) ]
CORR(s) 5
provided for each volume scan. In the typical rainfall sD(s) 3 sDref(s)
mode operation, NEXRAD radars collect a volume
scan every 4–6 min. To obtain 15-min time series, we first E[D(s,t) 2 E[D(s,t) ]] 3 E[Dref (s,t) 2 E[Dref(s,t) ]]
5 .
interpolated both variables to produce maps with 1-min sD(s) 3 sDref(s)
time resolution. We then aggregate the 1-min values to
the desired temporal resolution (15, 30, and 60 min). For CORR measures the degree of linear association be-
IPC, we applied linear interpolation, while for HHC we tween the radar and gauge measurements. The quanti-
used the nearest-neighbor interpolation method. ties E[X] and sX are the expected value and the
standard deviation of X, respectively. NB and CORR
c. Methods
are dimensionless, and SE is in millimeters or millime-
The goal of this paper is to assess improvements in ters per hour. The index s refers to a specific spatial
QPEs obtained by dual-polarization radars and to explore domain that can be a gauge location or a subbasins, and t
the potential of applying this dataset to urban hydrology to the time intervals. NB is a measure of the overall bias
even when ground data are not available. Therefore, we (B), which represents systematic average deviations of
evaluate radar rainfall estimates without removing sys- radar estimates with respect to the rain gauge estimation
tematic biases relative to rain gauges. Moreover, this over the spatial domain of interest (gauge location or
correction requires long series of data, which are not yet subbasin). SE is a measure of dispersion and indicates how
DECEMBER 2013 CUNHA ET AL. 1483
radar measurements differ from their expected value calculate statistical measures and hydrologic parameters
(gauge estimation). Since we do not remove the mean field that are relevant for flood simulation, including mean
bias, SE accounts for systematic and random errors. areal accumulated rainfall and mean areal maximum
CORR measures the degree of linear association be- rainfall intensity, for each of the nested subbasins, rang-
tween the radar and gauge measurements. ing in scale from 0.01 to 650 km2. We estimate the mean
Depending on the analyses, the variables D and Dref areal rainfall for each basin based on the weighted aver-
represent total storm accumulations (mm) or 15-min age of all radar pixels that are enclosed or intersected by
rain-rate values (mm h21) calculated based on different the basin. The weight is proportional to the area of the
rainfall datasets (gauge, SPR, or DPR) estimated over pixel that is contained by the basin. Note that even very
a point/pixel or over a certain spatial domain (subbasin). small basins can partially intersect more than one pixel.
The Dref refers to the gauge data, and the D to one of
the radar products. For the point analyses, the summa-
tions refers to each ith radar–gauge pair, yielding a 3. Results
sample size by event equal to the number of sites for total
a. Flood summary and storm totals
storm accumulations, and the number of sites times the
number of 15-min intervals for the rain-rate comparisons. The March, May, and September 2012 storm events
Rainfall is the main input for many hydrological exhibit contrasting characteristics in terms of rainfall
models; therefore, it is important to understand how distribution over the study region, magnitudes of rainfall
rainfall errors change across scales to be able to decide if rate, melting-layer position, and HT. In Fig. 2, we il-
a certain dataset is adequate for hydrological application lustrate the spatial structure of the storm total radar
(from hillslope to large-scale hydrology), and to better rainfall for a 160 km 3 120 km domain that contains the
understand uncertainties on the results of the hydro- Kansas City study region.
logical models (Seo et al. 2013). Previous studies have The 6–7 May 2012 storm (Fig. 2) was an organized
demonstrated that radar rainfall estimation errors de- thunderstorm system that produced damaging winds
crease with averaging in space and time (e.g., Knox and and hail. The storm total rainfall fields exhibit large
Anagnostou 2009; Seo and Krajewski 2010). This is ex- spatial variability associated with the structure and evo-
pected since averaging a large number of observations lution of the storm system. In Fig. 2, we present rainfall
reduces random uncertainties. This is relevant for the fields derived from both the DPR and SPR algorithms
hydrological community, since it means that even if for both the KEAX and KTWX radars. In addition to
errors are large for small scales (e.g., 15 min in time, the spatial variability in storm total rainfall for each of
1 km 3 1 km in space), there should be a scale for which the four rainfall fields, there are also striking contrasts
errors are in an acceptable range for flood applications. among the four radar rainfall fields. One of the central
In this study, we aggregate the information in time questions we want to address is whether the DPR rain-
(15–180 min) and in space (0.02–650 km2) and calculate fall estimates are superior to the SPR rainfall estimates.
statistical measures described in the previous sections Over the Kansas City study region, rain gauge storm
(NB, SD, and CORR). Because our focus is on pro- totals varied from 10 to 130 mm and peak rainfall in-
viding information about radar rainfall error scaling in tensities reached 150 mm h21 at 15-min time resolution.
a hydrologic context, we aggregate the information in During this event, the U.S. National Weather Service
space using basin boundaries, which are natural spatial issued numerous flood warnings for Johnson, Jackson,
domains in hydrology. We use the river networks anal- and Cass Counties that covered the Kansas City area. The
ysis tool CUENCAS [see description in Mantilla and local news channel (KSHB, channel 41) reported the
Gupta (2005) and Cunha et al. (2011)] to automatically occurrence of 1-in. hail in the southeast part of the city.
extract the river network and the basin boundaries The 19–20 March 2012 storm was a typical spring ex-
based on the U.S. Geological Survey’s (USGS) National tratropical system that produced moderately heavy
Elevation Dataset (NED) 30-m Digital Elevation Map rainfall (Fig. 3a) over time periods of 12–24 h. Over the
(DEM; Gesch et al. 2009). We interpolate 15-min rain course of the event there was a temperature drop of
gauge data using the inverse-distance-weighting method approximately 128C (218–98C) as a cold front passed
to generate spatially continuous rainfall maps with the through the region; the melting layer was low over the
same resolution as the radar maps. The interpolated entire duration of the storm. The KEAX radar detected
datasets are used as reference to evaluate radar rainfall a few periods (approximately 1% of the time, for se-
estimates. Due to the high density of gauges (see Fig. lected regions) with dry snow, while the measurements
1b), the interpolated maps provide accurate represen- performed by the KTWX radar included ice crystals, dry
tation of the rainfall spatial variability in the basin. We snow, wet snow, and graupel. These are hydrometeor
1484 WEATHER AND FORECASTING VOLUME 28
FIG. 2. Storm total accumulation field (mm) derived from (a) KEAX-SPR, (b) KEAX-DPR, (c) KTWX-SPR, and (d) KTWX-DPR
for the May 2012 storm.
types observed only in the melting layer. The storm total classified the radar target as ice crystal, dry snow, wet
measured by rain gauges varied from 45 to 70 mm over snow, or graupel during each storm. For the March event
the Kansas City study area. among the hydrometeor types are ice crystals, implying
The 31 August–1 September storm is typical of fall that the radar beam overshoots the precipitation. Dry
extratropical systems that produce high storm total rain- snow, wet snow, and graupel are identified when the radar
fall (Fig. 3b) over time periods of 12–24 h. The storm beam is observing the melting layer. For the March and
period was characterized by a cold front that became May events, the melting layer affects rainfall estimates. For
stationary over the region. In the beginning of the period, the September event there is only a brief period of time for
temperatures dropped from a maximum of 388C to 188C which the bright band affects the rainfall estimates.
and then remained constant for the rest of the period. The In Fig. 5, we present the frequency of HT for the May
event was characterized by low-intensity rainfall that event, for light–moderate rain, heavy rain, big drops,
lasted a long period, resulting in storm totals that varied and hail for the KEAX and KTWX radars. The maps
from 76 to 222 mm over the Kansas City study area. would be identical if both radars were observing the
In Fig. 4, we present maps of hydrometeor-type fre- same layer of the atmosphere and if the HT classification
quency for the KTWX radar for all events. The maps procedure was error free. The HT observed indicated that
represent the percentage of the time for which the radar the radar beam captures at least part of the atmosphere
DECEMBER 2013 CUNHA ET AL. 1485
FIG. 3. Storm total accumulation field (mm) derived from KEAX-DPR for the (a) March and (b) September 2012 storms.
below the melting layer. There are several reports of hail basin (less than 20 mm) and for regions where hail was
on the ground during the May event. Both radars detected detected. The KEAX-DPR rainfall estimates do not
hail mainly in the southern part of the study region. The exhibit systematic underestimation of storm totals. Al-
KTWX radar observed much more hail and more big though smaller in magnitude, bias still exists for the
drops than did the KEAX radar. This is expected because DPR rainfall estimates. For the March and May storms,
the KTWX radar is sampling higher layers of the atmo- the normalized bias shifts signs from SPRs to DPRs.
sphere. KEAX detected more heavy rain in the same areas KEAX-DPR overestimates the storm total accumulations.
where KTWX detected more hail. For the September storm, both KTWX-SPR and KEAX-
Figure 6 presents gauge versus radar scatterplots of DPR underestimate the storm total but KEAX-DPR has
storm total rainfall values for the March (green), May a smaller normalized bias.
(blue), and September (red) events for SPR (top) and KTWX-SPR also underestimates the storm total rain-
DPR (bottom) rainfall estimates and for the KEAX fall for the September event, for which no brightband
(left) and KTWX (right) radars. Table 2 summarizes contamination occurs even at the far ranges. It is likely
the normalized bias, standard errors, and correlation of that even at these far ranges the effect of biased Z–R re-
storm total radar rainfall estimates using rain gauge lationships can be seen when no brightband contamina-
observations as reference data. We also include the av- tion occurs. The opposite effect occurs for the events with
erage (of all gauges) of the percentage of time that each brightband contamination, for which KTWX-SPR over-
hydrometeor type was identified by KEAX and KTWX. estimates rainfall. The absence of a systematic pattern in
The values are calculated for each event and for all normalized bias will be discussed in the next section
the events together. Note that the KEAX-SPR always where we elaborate upon range effects on storm totals.
underestimates the storm totals for the March and The correlation between storm total rainfall from ra-
September events, and almost always for the May event. dar and rain gauge storm total accumulations is larger
Close-range biases for SPR are likely caused by a biased for all three events for the KEAX radar (Table 2). For
Z–R relationship that compensates for the typical the KTWX radar, DPR rainfall estimates have a larger
brightband contamination (Smith et al. 1996); the ab- correlation than SPR rainfall estimates for the May and
sence of brightband contamination in this case leads to September storms, but not for the March storm. The
underestimation at close range. For the May event, normalized bias is closer to 0 for the May and September
KEAX-SPR overestimates storm totals for some of the events for KEAX-DPR rainfall estimates. For KTWX-
gauges. Overestimation occurs for gauges with very low DPR, the normalized bias is closer to 0 for the May
storm totals in the southwest and in the north of the storm, but not for the March and September events.
1486 WEATHER AND FORECASTING VOLUME 28
FIG. 4. Frequency of (from left to right) hybrid hydrometeor classification for KTWX for HTs typical of the melting layer: (from top to
bottom) March, May, and September. These hydrometeors are not identified by the KEAX radar since it is located very close to the study
area and the radar beam is always below the melting layer.
These features are closely linked to range-dependent SPRs. As previously noted, SPRs use the traditional Z–R
sampling of the radars, as detailed below. relationship and the parameters of this relationship cannot
account for range-dependent effects of brightband con-
b. Range effects on storm totals
tamination. For the DPRs, rainfall is estimated based on
Figure 7 presents the normalized bias for storm total reflectivity, differential reflectivity, and hydrometeor class.
rainfall versus radar range for the KEAX (ranges up to The estimation algorithm generally results in lower bias;
72 km) and KTWX (ranges larger than 104 km) for SPRs however, for the March event, KEAX-DPR overestimates
(blue) and DPRs (red) for the March, May, and September storm totals relative to rain gauges. For the September
events. Each point represents NB values for a rain gauge event, KEAX-DPR underestimates the storm total in
location. We also include the moving-average lines esti- relation to the gauges. For the May event we observe
mated with a 10-km window to facilitate visual evaluation a large degree of scatter, likely resulting in part from the
of the data (blue and red lines). This representation of large variability of HT and storm totals as indicated by
range-dependent bias reflects some of the issues discussed Figs. 2 and 5, respectively. Independent of the scatter,
in the previous section, but also provides information about the regression lines show that the differences between
how radar uncertainties and improvements from SPR to KEAX-SPR and KEAX-DPR follow the same tendency
DPR rainfall estimates vary with range. as the ones observed for the other events.
KEAX storm total accumulations estimated by the For the KTWX radar the changes between the DPR
DPRs are always higher than the rainfall estimates by and SPR rainfall estimates are not always consistent. For
DECEMBER 2013 CUNHA ET AL. 1487
FIG. 5. Frequency of hybrid hydrometeor classification for (left)
KEAX and (right) KTWX for the May event: (from top to bottom) light–
moderate rain, heavy rain, big drop rain, and hail.
the September event, we observe almost no difference total estimates for ranges up to approximately 130 km but
between the KTWX-SPR and KTWX-DPR storm totals. the estimates deteriorated at larger ranges. For the May
For the March event, the differences between KTWX-SPR storm, we observed significant changes between KTWX-
and KTWX-DPR are large. KTWX-DPR improved storm DPR and KTWX-SPR for ranges up to 125 km, some1488 WEATHER AND FORECASTING VOLUME 28
FIG. 6. Scatterplot of gauge vs radar rainfall accumulation for the March (green), May (blue),
and September (red) events for the (top) SPR and (bottom) DPR datasets and for (left) KEAX
and (right) KTWX.
changes in the 125–140-km range, and almost no changes opposing patterns of behavior: for ranges less than 125 km,
for ranges larger than 140 km. KTWX-DPR exhibits superior estimation of storm total
Based on HHC frequency for the May and September rainfall compared to KTWX-SPR, and for ranges greater
storms (Table 2), the percentage of light–moderate rain than 125 km, KTWX-DPR storm total rainfall estimates
is not significantly different between KTWX and are worse than those of KTWX-SPR. In this section, we use
KEAX. The main difference is that KTWX detects pe- the 15-min rain-rate fields and the hydrometeor frequency
riods with graupel, dry snow, wet snow, and ice crystals— time series to investigate the cause of these contrasts.
hydrometeors typical of the melting layer—while KEAX In Figs. 8 and 9, we analyze the time series of rain
does not detect these HTs. Moreover, KTWX detects hail rate (gauge, SPR, and DPR) and HT frequency of two
more frequently than KEAX. For the March event, sig- gauges, located at distances of 105 km (gauge 1) and
nificant differences are observed between the hydrome- 152 km (gauge 2) from KTWX and 69 km (gauge 1) and
teor frequency observed by KEAX and by KTWX. For 47 km (gauge 2) from KEAX. Figures 8 and 9 present
that event, KTWX almost always samples above the the 15-min rain-rate time series from the rain gauges,
melting layer, resulting in large errors for KTWX-DPR KTWX-SPR, KEAX-DPR, KTWX-SPR, and KTWX-
rainfall fields. We will discuss this issue in the following DPR, as well as the HT frequencies for KEAX and
section. KTWX. These two examples demonstrate the lack of
consistency in improvements obtained by DPRs. For
c. Gauge–radar 15-min rain-rate intercomparison
gauge 1 (Fig. 8), KEAX overestimates the first rainfall
In the previous section we showed that differences peak for both the DPRs and SPRs, with the DPR esti-
between storm total accumulation based on KTWX-SPR mate being worse than that for the SPRs. For KTWX,
and KTWX-DPR for the March event are not always the DPRs overestimate and SPRs underestimate the
consistent and are range dependent. We observe two same peak, with the DPR estimation being considerablyDECEMBER 2013 CUNHA ET AL. 1489
TABLE 2. Storm total statistics for all rainfall datasets and events.
Mar May Sep All
Gauge Avg (mm) 47.40 49.59 126.77
Std dev (mm) 8.75 26.95 30.13
Coefficient of variation 0.18 0.54 0.24
NB KTWX-SPR 20.31 20.26 20.69 20.52
KEAX-DPR 0.31 0.21 20.19 0.00
KTWX-SPR 0.29 0.69 20.30 0.04
KTWX-DPR 0.70 0.56 20.34 0.07
CORR KTWX-SPR 0.77 0.73 0.61 0.46
KEAX-DPR 0.87 0.83 0.81 0.86
KTWX-SPR 0.77 0.85 0.38 0.53
KTWX-DPR 0.22 0.89 0.48 0.43
SE KTWX-SPR 15.95 22.55 91.02 55.39
KEAX-DPR 17.28 21.13 29.65 23.32
KTWX-SPR 21.24 41.18 46.86 38.02
KTWX-DPR 34.87 32.48 50.51 40.23
KEAX HT avg frequency % RA 28.87% 18.69% 54.88% 34.14%
% HR 0.32% 1.10% 0.00% 0.47%
% BD 0.29% 1.67% 0.21% 0.72%
HA 0.01 0.07 0.00 0.03
% GR, DS, WS, and IC 0.04% 0.00% 0.00% 0.01%
% others; no rain 70.47% 78.47% 44.91% 64.62%
KTWX HT avg frequency % RA 5.05% 17.58% 54.85% 25.83%
% HR 0.08% 1.09% 0.00% 0.39%
% BD 0.31% 1.70% 0.22% 0.74%
HA 0.05 0.34 0.00 0.13
% GR, DS, WS, and IC 41.98% 6.30% 0.72% 16.33%
% others; no rain 52.52% 73.00% 44.20% 56.57%
better than that of the SPRs. For the remaining time Analyses of 15-min rainfall rates (Fig. 10) indicate
period, with lighter rain, the KTWX-DPR and KTWX- that the DPR rainfall estimates for KEAX are signifi-
SPR rainfall estimates are similar for this gauge. For cantly better than the SPR rainfall estimates in terms of
gauge 2 (Fig. 9), KEAX-DPR provides a better esti- both bias and standard error. Figure 10 presents a scat-
mation of the peak than KEAX-SPR, while the KTWX- terplot of 15-min rain gauge rainfall rate versus 15-min
SPR and KTWX-DPR estimates of the first peak are radar rainfall rate for the radar bin containing the gauge.
practically the same. However, KTWX-DPR and KTWX- The points are color coded by event: green for the March
SPR provide distinct rainfall estimates for the period event, blue for the May event, and red for the September
with light rainfall. KTWX-SPR did not identify rainfall event. We use power-law functions to describe the dif-
for many periods, while KTWX-DPR overestimated ferences between radar and gauge rainfall as a function
rain. Note that KEAX observes mainly (almost 100%) of rain rate (deterministic error) (Ciach et al. 2007;
hydrometeors associated with rain (RA 1 HR 1 BD) for Villarini et al. 2009). We estimate the parameters of the
both gauges, while KTWX observes a mix of rainfall and power-law function for each rainfall product and event
melting layer HTs. When analyzing Figs. 8 and 9, we using the Levenberg–Marquardt algorithm. The improve-
should remember that tipping-bucket errors are large for ment in the 15-min radar rainfall estimates for KEAX-
low-intensity rainfall and that this instrument fails to DPR from KTWX-SPR is clear, while KTWX-SPR and
identify the time that the rainfall started and ended. KTWX-DPR show very similar rain-rate estimates.
Nevertheless, the volume of rainfall should be similar, In Table 3, we summarize the 15-min rainfall estima-
which is the case for the gauge 1 estimates at KTWX and tion properties for KEAX and KTWX and for SPR and
KEAX, but not for gauge 2, with KTWX-DPR over- DPR measurements. To remove the effects of very low
estimating and KTWX-SPR underestimating the rainfall. values (for which gauges are not very accurate), we
The difference in bias among the KTWX DPRs and SPRs calculate CORR, NB, and SE based on the periods for
for gauge 2 occurs because the DPR algorithm over- which gauges observed 15-min rain rates larger than
corrects rain rates for frozen hydrometeors, while the 4 mm h21. The correlation improves from SPRs to DPRs
standard Z–R relationship underestimates rain rates for all events and both radars, except the May storm
when applied for these hydrometeors. for KTWX. For the March storm, NB and SE are1490 WEATHER AND FORECASTING VOLUME 28
hydrometeors: rainfall only (RA, HR, BD), hail below the
melting layer (HA with RA, HR, and BD), hail above the
melting layer (HA, with GR, WS, DS, and IC), and
melting layer only (GR, WS, DS, and IC). DPR mea-
surements are better than the SPR measurements for all
cases, except when hail is observed above the melting
layer, for which the bias increases from 0.56 to 0.84 and
the correlation decreases from 0.52 to 0.17. This explains
why CORR decreases from KTWX-SPR to KTWX-DPR
for the May event, and points out the need to reevaluate
the radar rainfall estimation equation for this HT.
d. Time-scale dependency of errors
Radar rainfall errors are expected to decrease with
increasing temporal scales because uncertainties that
arise from the gauge–radar spatial sampling problem are
filtered out. Figure 11 shows how radar rainfall uncer-
tainty changes with temporal scales varying from 15 to
180 min for the March, May, and September events.
Based on this figure, radar measurement errors decrease
as we aggregate the observations in time. The rate of
decrease, however, is different for each event. For the
May and September events, CORR does not increase
significantly for scales greater than 1 h, while for the
March events CORR still increases as we aggregate the
observations up to 3 h. SE decreases up to scales of 3 h. As
SE is sensitive to the sample size, part of the decrease in SE
FIG. 7. Normalized bias as a function of range for the (top)
is caused by the sample size, which decreases as the tem-
March, (middle) May, and (bottom) September events and for SPR poral aggregation increases.
(blue) and DPR data (red). Ranges up to 70 km correspond to In terms of correlation (CORR), KEAX-DPR ex-
KEAX and ranges larger than 100 km to KTWX. hibits better performance for the March and September
events, while KTWX-DPR exhibits slightly better
performance for the May event. In terms of standard
considerably higher for the KEAX DPRs than for the error (SE), the DPRs provide better performance than
SPRs (from 0.50 to 20.28 for NB and from 9.42 to 5.75 the corresponding SPRs for the May and September
for SE; green dots in Fig. 6). For the May and September events. For the March event, KEAX-SPR exhibited the
events, KEAX-DPR performs better than KTWX-SPR. lowest SE among all radar–polarization combinations.
One important point to note is the reduction of spread
e. Basin-scale analyses of radar rainfall errors
that is probably the result of including HTs in the rainfall
estimation algorithm. For the KTWX radar, changes In this section, we investigate how rainfall uncer-
from SPR to DPR are not substantial for the May and tainties change across different spatial scales delineated
September events. Because we focus on large rain-rate by watersheds. We evaluate nested subbasins with drain-
values, for the March event, the DPRs provide better age area varying from 0.02 to 650 km2. We use CUENCAS
measurements than do the SPRs. For all cases the DPRs to obtain the river network and basin delineations [see
provide better estimates of maximum rainfall than do Mantilla and Gupta (2005) for details on the software
the SPRs, with the exception of the KEAX March event and the procedure], from the 30-m NED DEM provided
for which the maximum rainfall was overestimated by by the USGS. All analyses in this section are for the May
58%. event.
To understand why CORR decreases from KTWX- In Fig. 12, we present correlation and standard error
SPR to KTWX-DPR for the May event, we separate our estimates based on the 15-min mean areal rain rate for
sample size based on the HT and calculate the same the May event. We estimate the statistics using as ref-
statistics presented in Table 3. These statistics are pre- erence 15-min rain gauge rain rates that are spatially
sented in Table 4. We consider the following classes of interpolated to generate continuous spatial maps. TheDECEMBER 2013 CUNHA ET AL. 1491
FIG. 8. Rain-rate and HT time series for the March event for FIG. 9. As in Fig. 8, but for gauge 2 located at 47 and 152 km from
gauge 1 located 69 and 105 km from KEAX and KTWX, re- KEAX and KTWX, respectively.
spectively: (from top to the bottom) the 15-min rain rate based on
rain gauge, KEAX-SPR, and KEAX-DPR; HT frequency for
rainfall hydrometeors (RA 1 HR 1 BD) and HTs typical of the corner of each plot represent the NB-MAST for the
melting layer (GR 1 IC 1 DS 1 WS) for KEAX; and the last two 650 km2 watershed. NB-MAST equal to 0.0 indicate un-
panels, as in the top two, but for KTWX.
biased data, and values equal to 0.5 (20.5) mean that the
radar overestimated (underestimated) the mean areal
red dots represent the values for the KTWX-SPR for the storm total by 50%. As the previous analyses demon-
9357 subbasins. Both statistics present large variability strated, KTWX-SPR (KTWX-DPR) has a tendency to
for basins smaller than 1 km2. This is expected because underestimate (overestimate) rainfall for the May event.
the radar rainfall data resolution is 1 km2 and, thus, does For KTWX-DPR, many of the basins with area smaller
not provide accurate information for smaller scales. To than 1 km2 exhibited NB-MAST values larger than 1,
remove the effects of random variation and reveal un- which indicates an overestimation of more than 100%.
derlying trends, we include moving-average lines for For the outlet, NB-MAST is equal to 20.26 for KEAX-
KTWX-SPR (red line) and the remaining radar products. SPR and 0.66 for KTWX-SPR. Bias decreases for the
The plots show that correlation increases and standard DPR for both radars, but KTWX-DPR still shows large
error decreases with scale for all of the products. DPR bias (0.57). This demonstrates the need for more re-
products have better statistics than SPR products for the search to improve range-dependent properties of DPR
same radar. For the largest drainage area evaluated in this rainfall estimation.
study (650 km2), CORR is equal to 0.98 and SE to 1.7 mm The best product (KEAX-DPR) overestimates the
for the KEAX-DPR, with values of 0.98 and 2.2 mm, mean areal storm total by 23% at the outlet of the
respectively, for KTWX-DPR. basin (650 km2). Since rainfall is the major driving
In Fig. 13, we present the normalized bias for the force for flood modeling, radar rainfall estimates
mean areal storm total (NB-MAST) for the four radar should be bias corrected before being used as input
rainfall products for all nested watersheds. We include to flood simulation models. NB-MAST exhibits large
regression lines for all plots to assist in the evaluation variability for basins smaller than 10 km 2 for all
of the results. The numbers presented in the top-right products (e.g., 20.8 to 1.0 for KEAX-SPR). In future1492 WEATHER AND FORECASTING VOLUME 28
FIG. 10. Gauge–radar scatterplots of 15-min data for (left) KEAX and (right) KTWX with (top) SPR and
(bottom) DPR datasets for the March (green), May (blue), and September (red) events. Included are power laws to
describe the relationship between rain gauge and radar rain rate for each event.
studies, we will investigate how these uncertainties the study area allows us to investigate range effects on
propagate through hydrologic models and affect their SPR and DPR rainfall products. Radar rainfall esti-
flood predictions. mates are compared to tipping-bucket rain gauge ob-
servations from the dense gauge network in Kansas City.
We base our conclusions on analyses of three storm
4. Conclusions
events that occurred in March, May, and September of
In this study we compare radar rainfall fields based on 2012. The principal conclusions of these analyses are
SPR and DPR measurements from radar. We analyzed summarized below.
rainfall fields developed from the KEAX and KTWX 1) For a diverse range of storm types, DPR measure-
WSR-88Ds, and focus on a 60 km 3 60 km study region ments generally improve the quantitative estimation
in Kansas City with a network of 136 rain gauges. The of rainfall relative to SPR estimates. However,
distance from the KEAX radar to rain gauges in the improvements achieved by DPR are not consistent
Kansas City network spans from 23 to 72 km, while for all events or radars. The nature of the improve-
the distance from the KTWX radar extends from 104 to ment depends fundamentally on range-dependent
157 km. This difference in distance from the radars to sampling of the different hydrometeor types.DECEMBER 2013 CUNHA ET AL. 1493
TABLE 3. The 15-min rain-rate statistics for all rainfall datasets and events.
Statistics No. of time intervals with
Event Dataset NB CORR SE Rmax R.0 R.4 R . 25 R . 100
Event 1 (No. of sites 5 131; Gauge 54.1 9822 1174 36 0
No. of samples 5 201 203) KEAX-SPR 20.28 0.68 5.75 52.9 13 880 609 39 0
KEAX-DPR 0.5 0.82 9.42 85.6 15 646 1780 144 0
KTWX-SPR 20.24 0.52 6.44 44.6 10 551 1335 30 0
KTWX-DPR 20.02 0.65 5.2 53.5 15 631 3814 30 0
Event 2 (No. of sites 5 125; Gauge 147.5 4000 1078 254 7
No. of samples 5 201 205) KEAX-SPR 20.35 0.53 18.21 102.1 9399 783 197 1
KEAX-DPR 0.19 0.68 18.04 124.5 7079 1122 360 11
KTWX-SPR 0.49 0.72 20.17 103.8 6416 1660 506 13
KTWX-DPR 0.43 0.69 19.83 131.2 8178 1454 445 16
Event 1 (No. of sites 5 132; Gauge 33.1 12 004 6617 26 0
No. of samples 5 201 209) KEAX-SPR 20.7 0.45 6.91 13.8 13 553 1117 0 0
KEAX-DPR 20.19 0.78 3.19 35.7 14 601 5037 25 0
KTWX-SPR 20.43 0.3 5.92 24.9 12 207 4512 0 0
KTWX-DPR 20.42 0.46 5.18 29.6 14 732 4114 2 0
2) Improvements in rainfall estimation achieved by the 3) For the cases with no brightband contamination
use of DPR measurements are more pronounced for or hail, SPR rainfall predictions underestimate the
KEAX, the radar located close to the study area. For rainfall. This occurs for the March and September
the far-range KTWX radar, small improvements are events for the KEAX radar, and the September
observed for the May and September events through event for the KTWX radar. This is likely the result
the use of DPR measurements. For the March event, of using standard parameters for the Z–R relationships
rainfall estimates based on the DPR algorithm are that, on average, compensate for some brightband
better for a range interval of approximately 104– contamination. When mixed-phase hydrometeors
125 km, and worse (overestimation) at a range in- are not present, rainfall is underestimated using SPR
terval of approximately 125–156 km. This problem is reflectivity measurements. DPRs are able to detect
related to the detection of dry snow and ice crystals. settings in which rain is observed entirely below the
TABLE 4. The 15-min rain-rate statistics for the May event separated by HHC.
Statistics No. of points with
Sample size Dataset NB CORR SE Rmax R.0 R.4 R . 25 R . 100
Just rain (RA, HR, BD)
3374 Gauge 115.4 2338 830 195 4
KEAX-SPR 20.26 0.64 9.39 102.1 3068 573 159 1
KEAX-DPR 0.29 0.79 9.51 124.4 3374 866 292 7
4176 Gauge 66.6 2105 364 8 0
KTWX-SPR 0.88 0.6 4.2 91.7 3859 725 40 0
KTWX-DPR 0.55 0.61 2.78 27.9 4176 589 5 0
Hail below melting layer (HA with RA, HR, and BD)
69 Gauge 147.4 69 66 49 3
KEAX-SPR 20.5 0.56 33.02 86.9 68 53 24 0
KEAX-DPR 0.07 0.58 27.21 116.5 69 66 50 4
158 Gauge 147.4 157 148 96 7
KTWX-SPR 0.66 0.58 36.23 103.8 158 155 138 12
KTWX-DPR 0.44 0.51 32.63 118.0 158 155 127 12
Hail above melting layer (HA with GR, WS, DS, and IC)
37 Gauge 85.9 37 34 15 0
KTWX-SPR 0.56 0.52 25.18 101.3 37 37 31 1
KTWX-DPR 0.84 0.17 32.25 88.2 37 37 36 0
Melting layer (GR, WS, DS, and IC)
349 Gauge 112.0 252 62 16 1
KTWX-SPR 0.61 0.8 11.32 103.7 236 128 31 3
KTWX-DPR 0.68 0.85 9.38 86.8 349 101 36 01494 WEATHER AND FORECASTING VOLUME 28
FIG. 11. Error statistics (radar products compared to gauge rainfall data) with respect to temporal scale for (top to
bottom) the March, May, and September events: (left) CORR and (right) SE.
bright band. For these situations, rainfall is estimated observations of the May storm, we did not find signif-
as a function of reflectivity and differential reflectivity. icant improvements from DPR measurements.
The addition of differential reflectivity to the rainfall 5) Radar–rain gauge comparisons demonstrate the large
estimation process has a significant positive impact on bias for the KEAX-SPR estimates, and the ability of
the values estimated for the KEAX radar. the DPR algorithm to correct this bias. The normal-
4) DPRs improved the estimation of rainfall when hail ized bias improved from 20.52 to 0.01. KTWX-SPR
occurred below the top of the melting layer, as was rain rates are characterized by low bias but large
the case for the Kansas City observations during the random error, which is slightly improved by the DPR
May event. When hail mixed with rain is detected measurements (decrease in standard error from 3.62
above the melting layer, as was the case for the KTWX to 3.28).DECEMBER 2013 CUNHA ET AL. 1495
FIG. 12. Error statistics for 15-min rain rates with respect to spatial scale for nested subwatersheds in the study area (May event): (left)
CORR and (right) SE.
6) Comparisons of KEAX and KTWX rainfall esti- the radar detects only raindrops, but inconsistent
mates demonstrate that operational DPR algorithms otherwise. This might arise from the fact that QPE
reduce range effects, relative to SPR rainfall esti- algorithms were developed based on data collected
mates. In general, improvements are substantial when in Oklahoma during the warm season, when snow is
FIG. 13. Mean areal storm total normalized biases as a function of spatial scale for nested subwatersheds in the
study area (May event): (top left) KEAX SPR and (top right) DPR; (bottom) as in (top), but for KTWX. The
number at the top right of each plot represents the NB for the 650 km2 watershed.1496 WEATHER AND FORECASTING VOLUME 28
uncommon in the atmospheric column. Uncer- Summers endowment. This research was supported
tainties in DPR QPEs can also be caused by in- by the Willis Research Network, the National Science
strument miscalibration (Ryzhkov et al. 2005a). We Foundation (Grant CBET-1058027), and the NOAA
cannot further investigate this source of error since we Cooperative Institute for Climate Science.
do not have information about the calibration pro-
cedures performed in KEAX and KTWX, or long
REFERENCES
series of data to statistically quantify errors Additional
research is needed to address the range-dependent Austin, P. M., 1987: Relation between measured radar reflectivity
error structure of DPR rainfall estimates. and surface rainfall. Mon. Wea. Rev., 115, 1053–1070.
Baeck, M. L., and J. A. Smith, 1998: Estimation of heavy rainfall by
7) We demonstrate that the error structure of DPR
the WSR-88D. Wea. Forecasting, 13, 416–436.
QPEs varies considerably from event to event. This is Benjamin, S. G., 1989: An isentropic mesoa-scale analysis system
expected since different radar rainfall algorithms and its sensitivity to aircraft and surface observations. Mon.
are used depending on the type of hydrometeor Wea. Rev., 117, 1586–1603.
observed. Empirical models that characterize rainfall Brandes, E. A., and K. Ikeda, 2004: Freezing-level estimation with
polarimetric radar. J. Appl. Meteor., 43, 1541–1553.
error structure of DPRs (e.g., Villarini et al. 2009)
Carpenter, T. M., and K. P. Georgakakos, 2004: Impacts of para-
should explicitly account for range-dependent sam- metric and radar rainfall uncertainty on the ensemble stream-
pling and the effects of HTs. flow simulations of a distributed hydrologic model. J. Hydrol.,
8) As we aggregate the observations in time, the corre- 298, 202–221, doi:10.1016/j.jhydrol.2004.03.036.
lation increases and the standard error decreases Chandrasekar, V., V. N. Bringi, N. Balakrishnan, and D. S. Zrnic,
1990: Error structure of multiparameter radar and surface
for all of the rainfall fields. The correlation did not
measurements of rainfall. Part III: Specific differential phase.
improve significantly for aggregation times larger than J. Atmos. Oceanic Technol., 7, 621–629.
1 h for the May and September events, while for the Ciach, G. J., 2003: Local random errors in tipping-bucket rain gauge
March event the correlation continues to increase for measurements. J. Atmos. Oceanic Technol., 20, 752–759.
aggregation time up to 3 h (maximum aggregation ——, and W. F. Krajewski, 1999: On the estimation of radar rainfall
error variance. Adv. Water Resour., 22, 585–595, doi:10.1016/
time adopted in the study). For the same radar, DPR
S0309-1708(98)00043-8.
rainfall fields provide higher correlations for all events ——, ——, and G. Villarini, 2007: Product-error-driven uncer-
compared to the SPR fields. We also analyze un- tainty model for probabilistic quantitative precipitation esti-
certainty as a function of spatial scale. We use basin mation with NEXRAD data. J. Hydrometeor., 8, 1325–1347.
boundaries to delineate the spatial domain and ag- Cunha, L. K., W. F. Krajewski, and R. Mantilla, 2011: A framework
for flood risk assessment under nonstationary conditions or in
gregate the information in space. The analyses dem-
the absence of historical data. J. Flood Risk Manage., 4, 3–22.
onstrate that for basins up to approximately 20 km2 ——, P. V. Mandapaka, W. F. Krajewski, R. Mantilla, and A. A.
the differences in total accumulations obtained by Bradley, 2012: Impact of radar-rainfall error structure on es-
gauges and radar exhibit large variability and reach timated flood magnitude across scales: An investigation based
values of up to 300% for the KTWX-SPR rainfall on a parsimonious distributed hydrological model. Water Re-
sour. Res., 48, W10515, doi:10.1029/2012WR012138.
fields. Differences for larger basins decrease but are
Einfalt, T., K. Arnbjerg-Nielsen, C. Golz, N.-E. Jensen,
still significant. For the largest basin (;650 km2) and M. Quirmbach, G. Vaes, and B. Vieux, 2004: Towards a
the best rainfall estimation, storm total radar rainfall is roadmap for use of radar rainfall data in urban drainage.
overestimated by 23%. This implies that real-time and J. Hydrol., 299, 186–202.
retrospective gauge bias correction, as generally ap- Fabry, F., G. L. Austin, and D. Tees, 1992: The accuracy of rainfall
estimates by radar as a function of range. Quart. J. Roy. Me-
plied to SPR, are likely to be useful for DPRs as well.
teor. Soc., 118, 435–453.
Fo, A. J. P., K. C. Crawford, and C. L. Hartzell, 1998: Improving WSR-
This paper is an early attempt to evaluate the system 88D hourly rainfall estimates. Wea. Forecasting, 13, 1016–1028.
recently implemented by the NWS to estimate rainfall Fulton, R. A., J. P. Breidenbach, D.-J. Seo, D. A. Miller, and
based on dual-polarization radars. Future work should T. O’Bannon, 1998: The WSR-88D rainfall algorithm. Wea.
Forecasting, 13, 377–395.
include more events, allowing the use of more rigorous
Germann, U., G. Galli, M. Boscacci, and M. Bolliger, 2006: Radar
statistical approaches to quantify radar rainfall errors as precipitation measurement in a mountainous region. Quart.
a function of range and hydrometeors types. J. Roy. Meteor. Soc., 132, 1669–1692.
Gesch, D., G. Evans, J. Mauck, J. Hutchinson, and W. J. Carswell
Acknowledgments. The authors would like to ac- Jr., 2009. The National Map—Elevation. U.S. Geological
Survey Fact Sheet 2009-3053, 4 pp. [Available online at http://
knowledge the city of Overland Park Kansas, and spe-
pubs.usgs.gov/fs/2009/3053/.]
cially Mike Ross, for providing helpful comments about Giangrande, S. E., and A. V. Ryzhkov, 2008: Estimation of rainfall
the rain gauge network. WFK acknowledges support based on the results of polarimetric echo classification. J. Appl.
from the Iowa Flood Center and the Rose and Joseph Meteor. Climatol., 47, 2445–2460.You can also read
